def genData_Test():
#dataHistFile = open('dat.pkl', 'r+b')
#dataHist = pickle.load(dataHistFile)
# dataFileNumber = dataHist['data_file_number'] + 1
# dataFile = open('data/dat_' + dataFileNumber + '.csv', 'a')
dataFile = open('test_data/dat_0.csv', 'a')
csvWriter = csv.writer(dataFile)
# date = dateHist['last_updated']
date = datetime.datetime(2018,4,12).date()
endDate = datetime.date.today()
while(date <= endDate):
print 'Checking data for ' + date.strftime('%Y-%m-%d')
day = date.weekday()
if(day == 4 or day == 5):
date += datetime.timedelta(days=1)
continue
fname = date.strftime('%Y-%m-%d')
print("File Name: " + fname)
file = open('test_data/news/' + fname + '.csv')
csv_file = csv.reader(file)
for row in csv_file:
stockdata = getStockData_Test(row[0], date)
if(stockdata == -1):
continue
print("Origin String: %s" % row[1])
filtered_string = filterString(row[1])
filtered_string = filtered_string.decode("ascii", errors="ignore").encode()
print("Filtered String: %s" % filtered_string)
try:
sentdata = sentiment.analyzeText(filtered_string)
except:
enchant_string = filterByEnchant(filtered_string)
print("Enchanted String: %s" % enchant_string)
try:
sentdata = sentiment.analyzeText(enchant_string)
except:
sentdata = sentiment.analyzeText(row[0])
data = []
data.extend((row[0], date.timetuple().tm_yday))
data.extend((sentdata.score, sentdata.magnitude))
# row 0: adj close, row 1: close, row 3: high, row 4: low, row 5: open, row 6: volume
data.extend((stockdata[5],stockdata[1],stockdata[0],stockdata[3],stockdata[4],stockdata[6]))
# row 0: symbol, row 1: date, row 2: sentiment score, row 3: sentiment magnitude, row 4: open, row 5: close, row 6: adj close, row 7: high, row 8: low, row 9: volume
csvWriter.writerow(data)
date += datetime.timedelta(days=1)
def get_user_response():
print("rendering response\n")
input = request.args.get('text')
print("outputting")
out = queryAPIForRecipe(sentiment.getIDs(sentiment.analyzeText(input), 0)).split(" - ")
return render_template("recipes.html", data=out[0], img=out[1], time=out[2])
def genData():
#dataHistFile = open('dat.pkl', 'r+b')
#dataHist = pickle.load(dataHistFile)
#dataFileNumber = dataHist['data_file_number'] + 1
#dataFile = open('data/dat_' + dataFileNumber + '.csv', 'a')
dataFile = open('data/dat_9.csv', 'a')
csvWriter = csv.writer(dataFile)
#date = dateHist['last_updated']
#endDate = datetime.date.today()
date = datetime.datetime.strptime('27012014', "%d%m%Y").date()
endDate = datetime.datetime.strptime('01012017', "%d%m%Y").date()
c = csv.reader(open('dict.csv','r'))
dic = {}
for rw in c:
dic[rw[1]] = rw[0]
while(date < endDate):
#print 'Checking data for ' + date.strftime('%Y-%m-%d')
day = date.weekday()
if(day == 4 or day == 5):
date += datetime.timedelta(days=1)
continue
fname = date.strftime('%Y-%m-%d')
file = open('data/news1/' + fname + '.csv')
csv_file = csv.reader(file)
for row in csv_file:
stockdata = getStockData(dic[row[0]], date)
if(stockdata == -1):
continue
t = ''
news = t.join(row[1:])
#print (news)
pos, neg = sentiment.analyzeText(news)
data = []
data.extend((row[0], date.timetuple().tm_yday))
data.append(pos)
data.append(neg)
data.extend(stockdata)
csvWriter.writerow(data)
date += datetime.timedelta(days=1)
def genData():
dataHistFile = open('dat.pkl', 'r+b')
dataHist = pickle.load(dataHistFile)
dataFileNumber = dataHist['data_file_number'] + 1
dataFile = open('data/dat_' + dataFileNumber + '.csv', 'a')
csvWriter = csv.writer(dataFile)
date = dateHist['last_updated']
endDate = datetime.date.today()
while (date < endDate):
print 'Checking data for ' + date.strftime('%Y-%m-%d')
day = date.weekday()
if (day == 4 or day == 5):
date += datetime.timedelta(days=1)
continue
fname = date.strftime('%Y-%m-%d')
file = open('data/news/' + fname + '.csv')
csv_file = csv.reader(file)
for row in csv_file:
stockdata = getStockData(row[0], date)
if (stockdata == -1):
continue
sentdata = sentiment.analyzeText(row[1])
data = []
data.extend((row[0], date.timetuple().tm_yday))
data.extend((sentdata.score, sentdata.magnitude))
data.extend(stockdata)
csvWriter.writerow(data)
date += datetime.timedelta(days=1)
dataHist['data_file_number'] = dataFileNumber
dataHist['last_updated'] = endDate
dataHistFile.seek(0)
pickle.dump(dataHist, dataHistFile, protocol=pickle.HIGHEST_PROTOCOL)
dataHistFile.close()
def genData():
#dataHistFile = open('dat.pkl', 'r+b')
#dataHist = pickle.load(dataHistFile)
# dataFileNumber = dataHist['data_file_number'] + 1
# dataFile = open('data/dat_' + dataFileNumber + '.csv', 'a')
dataFile = open('data/dat_1.csv', 'a')
csvWriter = csv.writer(dataFile)
# date = dateHist['last_updated']
date = datetime.datetime(2018,01,02).date()
endDate = datetime.date.today()
while(date < endDate):
print 'Checking data for ' + date.strftime('%Y-%m-%d')
day = date.weekday()
if(day == 4 or day == 5):
date += datetime.timedelta(days=1)
continue
fname = date.strftime('%Y-%m-%d')
file = open('data/news/' + fname + '.csv')
csv_file = csv.reader(file)
for row in csv_file:
stockdata = getStockData(row[0], date)
if(stockdata == -1):
continue
sentdata = sentiment.analyzeText(row[1])
data = []
data.extend((row[0], date.timetuple().tm_yday))
data.extend((sentdata.score, sentdata.magnitude))
data.extend((stockdata[5],stockdata[1],stockdata[0],stockdata[3],stockdata[4],stockdata[6]))
csvWriter.writerow(data)
date += datetime.timedelta(days=1)
def prediction(dt):
tf.reset_default_graph()
session = tf.Session()
#init = tf.global_variables_initializer()
#session.run(init)
model_path = os.getcwd() + '/model/lin-model.ckpt'
meta_path = os.getcwd() + '/model/lin-model.ckpt.meta'
#saver = tf.train.Saver()
saver = tf.train.import_meta_graph(meta_path)
saver.restore(session, model_path)
# p_test = sess.run(output_layer, feed_dict={stock_data: p_x, stock_price: p_y, keep_prob_input: 1.0,
# keep_prob_hidden: 1.0})
graph = tf.get_default_graph()
# output_layer = tf.get_collection('output_layer')[0]
# stock_data = tf.get_collection('placeholder')[0]
# stock_price = tf.get_collection('placeholder')[1]
stock_data = graph.get_tensor_by_name('stock_data:0')
opening_price = graph.get_tensor_by_name('opening_price:0')
stock_price = graph.get_tensor_by_name('stock_price:0')
W = tf.get_variable("Weight", [3, 1], dtype=np.float32)
y = graph.get_tensor_by_name('operator:0')
#keep_prob_input = graph.get_tensor_by_name('keep_prob_input:0')
#keep_prob_hidden = graph.get_tensor_by_name('keep_prob_hidden:0')
#output_layer = graph.get_tensor_by_name('output_layer:0')
today = datetime.datetime.today()
t_news_file = open('test_data/news/' + dt.strftime('%Y-%m-%d') + '.csv')
print('Collecting Test Data for %s' %
(dt + datetime.timedelta(days=1)).strftime('%Y-%m-%d'))
p_single_data = []
p_next_data = []
p_csv_file = csv.reader(t_news_file)
for row in p_csv_file:
p_stockdata = gendata.getStockData_Test(row[0], dt)
if (p_stockdata == -1):
continue
# sentdata = sentiment.analyzeText(row[1])
print("Origin String: %s" % row[1])
filtered_string = gendata.filterString(row[1])
filtered_string = filtered_string.decode("ascii",
errors="ignore").encode()
print("Filtered String: %s" % filtered_string)
try:
sentdata = sentiment.analyzeText(filtered_string)
except:
enchant_string = gendata.filterByEnchant(filtered_string)
print("Enchanted String: %s" % enchant_string)
try:
sentdata = sentiment.analyzeText(enchant_string)
except:
sentdata = sentiment.analyzeText(row[0])
data = []
data.extend((row[0], dt.timetuple().tm_yday))
data.extend((sentdata.score, sentdata.magnitude))
data.extend((p_stockdata[5], p_stockdata[1], p_stockdata[0],
p_stockdata[3], p_stockdata[4], p_stockdata[6]))
p_single_data.append(data)
if (dt < today):
dtnext = dt + datetime.timedelta(days=1)
p_stocknextday = gendata.getStockData_Test(row[0], dtnext)
if (p_stocknextday is not -1):
datanext = []
datanext.extend((row[0], dtnext.timetuple().tm_yday))
datanext.extend((sentdata.score, sentdata.magnitude))
datanext.extend(
(p_stocknextday[5], p_stocknextday[1], p_stocknextday[0],
p_stocknextday[3], p_stocknextday[4], p_stocknextday[6]))
p_next_data.append(datanext)
if len(p_single_data) > 0:
p_data = np.array(p_single_data)
# print('Test Data collected for %s data:' % (dt + datetime.timedelta(days=1)).strftime('%Y-%m-%d') + ', '.join(
# str(c) for c in p_data))
print('Test Data collected for %s data:' %
(dt + datetime.timedelta(days=1)).strftime('%Y-%m-%d'))
p_origin = p_data
p_data = preprocessData(p_data)
params = p_data.shape[1] - 1
p_x = p_data[:, 0:params]
p_y = p_data[:, params].reshape(-1, 1)
p_opening_price = p_x[:, params - 1].reshape(-1, 1)
shape = p_x.shape
print(shape)
p_test = session.run(y,
feed_dict={
stock_data: p_x,
opening_price: p_opening_price,
stock_price: p_y
})
origin_test = p_origin[:, 2:6]
for i in range(origin_test.shape[0]):
norm = np.linalg.norm(origin_test[i, :], axis=0)
p_test[i, 0] = p_test[i, 0] * norm
result_date = dt + datetime.timedelta(days=1)
print(
'Predict Data for %s :' % result_date.strftime('%Y-%m-%d') +
', '.join('Stock %s :%s' % (d[0], str(c))
for d, c in zip(p_single_data, p_test)))
p_compare_data = []
if (dt < today and len(p_next_data) > 0):
p_compare_data = p_single_data
else:
p_compare_data = p_single_data
s_name = []
s_close = []
s_adjclose = []
s_low = []
s_high = []
s_volume = []
s_open = []
s_next_close = []
for i in range(len(p_compare_data)):
s_name.append(p_compare_data[i][0])
s_close.append(p_compare_data[i][5])
s_adjclose.append(p_compare_data[i][6])
s_low.append(p_compare_data[i][8])
s_high.append(p_compare_data[i][7])
s_volume.append(p_compare_data[i][9])
s_open.append(p_compare_data[i][4])
if dt < today:
s_next_close.append(p_next_data[i][5])
if (dt == today):
p_result = pd.DataFrame({
'symbol': s_name,
'prediction': p_test[:, 0],
'previous open': s_open,
'previous close': s_close,
'previous high': s_high,
'previous low': s_low,
'previous volume': s_volume
})
else:
p_result = pd.DataFrame({
'symbol': s_name,
'prediction': p_test[:, 0],
'open': s_open,
'close': s_close,
'next close': s_next_close,
# 'prev close': s_adjclose,
'high': s_high,
'low': s_low,
'volume': s_volume
})
csv_name = 'test_data/line-result/result-%s.csv' % (result_date.date())
# file_result = open('data/result/' + csv_name, 'rb+')
p_result.to_csv(csv_name)
def learn_test(data):
data = preprocessData(data)
num_params = data.shape[1] - 1
X = data[:, 0:num_params]
Y = data[:, num_params].reshape(-1, 1)
# Split the data into training and testing sets (70/30)
train_X, test_X, train_Y, test_Y = cross_validation.train_test_split(
X, Y, test_size=0.30)
train_opening_price = train_X[:, num_params - 1].reshape(-1, 1)
test_opening_price = test_X[:, num_params - 1].reshape(-1, 1)
# Get the initial stock prices for computing the relative cost
stock_data = tf.placeholder(tf.float32, [None, num_params])
opening_price = tf.placeholder(tf.float32, [None, 1])
stock_price = tf.placeholder(tf.float32, [None, 1])
# Number of neurons in the hidden layer
n_hidden_1 = 3
n_hidden_2 = 3
weights = {'out': tf.Variable(tf.random_normal([n_hidden_2, 1]))}
biases = {'out': tf.Variable(tf.random_normal([1]))}
# Implement dropout to reduce overfitting
keep_prob_input = tf.placeholder(tf.float32)
keep_prob_hidden = tf.placeholder(tf.float32)
# Hidden layers
input_dropout = tf.nn.dropout(stock_data, keep_prob_input)
layer_1 = slim.fully_connected(input_dropout,
n_hidden_1,
biases_initializer=None,
activation_fn=tf.nn.relu)
layer_1_dropout = tf.nn.dropout(layer_1, keep_prob_hidden)
layer_2 = slim.fully_connected(input_dropout,
n_hidden_1,
biases_initializer=None,
activation_fn=tf.nn.relu)
layer_2_dropout = tf.nn.dropout(layer_2, keep_prob_hidden)
output_layer = tf.add(tf.matmul(layer_2_dropout, weights['out']),
biases['out'])
learning_rate = 1e-4
cost_function = tf.reduce_mean(
tf.pow(tf.div(tf.subtract(stock_price, output_layer), opening_price),
2))
optimizer = tf.train.AdamOptimizer(learning_rate).minimize(cost_function)
last_cost = 0
tolerance = 1e-6
epochs = 1
max_epochs = 1e6
sess = tf.Session()
with sess.as_default():
init = tf.global_variables_initializer()
sess.run(init)
while True:
sess.run(optimizer,
feed_dict={
stock_data: train_X,
opening_price: train_opening_price,
stock_price: train_Y,
keep_prob_input: 0.8,
keep_prob_hidden: 0.5
})
if epochs % 100 == 0:
cost = sess.run(cost_function,
feed_dict={
stock_data: train_X,
opening_price: train_opening_price,
stock_price: train_Y,
keep_prob_input: 0.8,
keep_prob_hidden: 0.5
})
print "Epoch: %d: Error: %f" % (epochs, cost)
if abs(cost - last_cost) <= tolerance or epochs > max_epochs:
print "Converged."
break
last_cost = cost
epochs += 1
print "Test error: ", sess.run(cost_function,
feed_dict={
stock_data: test_X,
opening_price: test_opening_price,
stock_price: test_Y,
keep_prob_input: 1.0,
keep_prob_hidden: 1.0
})
test_results = sess.run(output_layer,
feed_dict={
stock_data: test_X,
stock_price: test_Y,
keep_prob_input: 1.0,
keep_prob_hidden: 1.0
})
avg_perc_error = 0
max_perc_error = 0
mei = 0
for i in range(len(test_Y)):
actual_change = abs(test_Y[i][0] -
test_X[i][num_params - 1]) / test_X[i][num_params -
1]
predicted_change = abs(test_results[i][0] - test_X[i][num_params - 1]
) / test_X[i][num_params - 1]
delta = abs(actual_change - predicted_change)
avg_perc_error = avg_perc_error + delta
if delta > max_perc_error:
max_perc_error = delta
mei = i
avg_perc_error = (avg_perc_error * 100) / len(test_Y)
max_perc_error *= 100
print "Maximum percentage error: %f\nAverage percentage error: %f\n" % (
max_perc_error, avg_perc_error)
test_dates = []
#p_f = open(FNAME, 'r')
# stocks = p_f.readLines()
t_date = datetime.datetime(2018, 03, 01)
for i in range(4):
test_dates.append(t_date)
t_date += datetime.timedelta(days=1)
t_date = datetime.datetime(2018, 03, 16)
for i in range(3):
test_dates.append(t_date)
t_date += datetime.timedelta(days=1)
print('Prediction Test Dates:' +
', '.join(d.strftime('%Y-%m-%d') for d in test_dates))
precit_data = []
for dt in test_dates:
t_news_file = open('test_data/news/' + dt.strftime('%Y-%m-%d') +
'.csv')
print('Collecting Test Data for %s' %
(dt + datetime.timedelta(days=1)).strftime('%Y-%m-%d'))
p_single_data = []
p_csv_file = csv.reader(t_news_file)
for row in p_csv_file:
p_stockdata = gendata.getStockData_Test(row[0], dt)
if (p_stockdata == -1):
continue
sentdata = sentiment.analyzeText(row[1])
data = []
data.extend((row[0], dt.timetuple().tm_yday))
data.extend((sentdata.score, sentdata.magnitude))
data.extend((p_stockdata[5], p_stockdata[1], p_stockdata[0],
p_stockdata[3], p_stockdata[4], p_stockdata[6]))
p_single_data.append(data)
if len(p_single_data) > 0:
p_data = np.array(p_single_data)
print(
'Test Data collected for %s data:' %
(dt + datetime.timedelta(days=1)).strftime('%Y-%m-%d') +
', '.join(str(c) for c in p_data))
p_origin = p_data
p_data = normalizeInput(p_data)
params = p_data.shape[1] - 1
p_x = p_data[:, 0:params]
p_y = p_data[:, params].reshape(-1, 1)
p_test = sess.run(output_layer,
feed_dict={
stock_data: p_x,
stock_price: p_y,
keep_prob_input: 1.0,
keep_prob_hidden: 1.0
})
orign_test = p_origin[:, 2:6]
for i in range(orign_test.shape[0]):
norm = np.linalg.norm(orign_test[i, :], axis=0)
p_test[i, 0] = p_test[i, 0] * norm
result_date = dt + datetime.timedelta(days=1)
print(
'Predict Data for %s :' % result_date.strftime('%Y-%m-%d') +
', '.join('Stock %s :%s' % (d[0], str(c))
for d, c in zip(p_single_data, p_test)))
s_name = []
s_close = []
s_adjclose = []
s_low = []
s_high = []
s_volume = []
s_open = []
for i in range(len(p_single_data)):
s_name.append(p_single_data[i][0])
s_close.append(p_single_data[i][5])
s_adjclose.append(p_single_data[i][6])
s_low.append(p_single_data[i][8])
s_high.append(p_single_data[i][7])
s_volume.append(p_single_data[i][9])
s_open.append(p_single_data[i][4])
p_result = pd.DataFrame({
'symbol': s_name,
'prediction': p_test[:, 0],
'open': s_open,
'close': s_close,
'prev close': s_adjclose,
'high': s_high,
'low': s_low,
'volume': s_volume
})
csv_name = 'test_data/result/result-%s.csv' % result_date.date()
#file_result = open('data/result/' + csv_name, 'rb+')
p_result.to_csv(csv_name)
def test_prediction(type):
test_dates = []
# t_date = datetime.datetime(2018, 03, 01)
# for i in range(4):
# test_dates.append(t_date)
# t_date += datetime.timedelta(days=1)
t_date = datetime.datetime(2018, 03, 16)
for i in range(3):
test_dates.append(t_date)
t_date += datetime.timedelta(days=1)
with tf.Session() as session:
init = tf.global_variables_initializer()
session.run(init)
model_path = os.getcwd() + '/model/model_' + str(type) + '.ckpt'
meta_path = os.getcwd() + '/model/model_' + str(type) + '.ckpt.meta'
saver = tf.train.import_meta_graph(meta_path)
saver.restore(session, model_path)
# p_test = sess.run(output_layer, feed_dict={stock_data: p_x, stock_price: p_y, keep_prob_input: 1.0,
# keep_prob_hidden: 1.0})
graph = tf.get_default_graph()
# output_layer = tf.get_collection('output_layer')[0]
# stock_data = tf.get_collection('placeholder')[0]
# stock_price = tf.get_collection('placeholder')[1]
stock_data = graph.get_tensor_by_name('stock_data:0')
opening_price = graph.get_tensor_by_name('opening_price:0')
stock_price = graph.get_tensor_by_name('stock_price:0')
keep_prob_input = graph.get_tensor_by_name('keep_prob_input:0')
keep_prob_hidden = graph.get_tensor_by_name('keep_prob_hidden:0')
output_layer = graph.get_tensor_by_name('output_layer:0')
print('Prediction Test Dates:' +
', '.join(d.strftime('%Y-%m-%d') for d in test_dates))
precit_data = []
for dt in test_dates:
t_news_file = open('test_data/news/' + dt.strftime('%Y-%m-%d') +
'.csv')
print('Collecting Test Data for %s' %
(dt + datetime.timedelta(days=1)).strftime('%Y-%m-%d'))
p_single_data = []
p_csv_file = csv.reader(t_news_file)
for row in p_csv_file:
p_stockdata = gendata.getStockData_Test(row[0], dt)
if (p_stockdata == -1):
continue
sentdata = sentiment.analyzeText(row[1])
data = []
data.extend((row[0], dt.timetuple().tm_yday))
data.extend((sentdata.score, sentdata.magnitude))
data.extend((p_stockdata[5], p_stockdata[1], p_stockdata[0],
p_stockdata[3], p_stockdata[4], p_stockdata[6]))
p_single_data.append(data)
if len(p_single_data) > 0:
p_data = np.array(p_single_data)
print(
'Test Data collected for %s data:' %
(dt + datetime.timedelta(days=1)).strftime('%Y-%m-%d') +
', '.join(str(c) for c in p_data))
p_origin = p_data
p_data = normalizeInput(p_data, type)
if type == 0:
params = p_data.shape[1] - 1
else:
params = p_data.shape[1] - 2
p_x = p_data[:, 0:params]
p_y = p_data[:, params].reshape(-1, 1)
p_test = session.run(output_layer,
feed_dict={
stock_data: p_x,
stock_price: p_y,
keep_prob_input: 1.0,
keep_prob_hidden: 1.0
})
origin_test = None
if type == 0:
origin_test = p_origin[:, 2:6]
else:
origin_test = p_origin[:, 2:7]
for i in range(origin_test.shape[0]):
norm = np.linalg.norm(origin_test[i, :], axis=0)
p_test[i, 0] = p_test[i, 0] * norm
result_date = dt + datetime.timedelta(days=1)
print(
'Predict Data for %s :' % result_date.strftime('%Y-%m-%d')
+ ', '.join('Stock %s :%s' % (d[0], str(c))
for d, c in zip(p_single_data, p_test)))
s_name = []
s_close = []
s_adjclose = []
s_low = []
s_high = []
s_volume = []
s_open = []
for i in range(len(p_single_data)):
s_name.append(p_single_data[i][0])
s_close.append(p_single_data[i][5])
s_adjclose.append(p_single_data[i][6])
s_low.append(p_single_data[i][8])
s_high.append(p_single_data[i][7])
s_volume.append(p_single_data[i][9])
s_open.append(p_single_data[i][4])
p_result = pd.DataFrame({
'symbol': s_name,
'prediction': p_test[:, 0],
'open': s_open,
'close': s_close,
# 'prev close': s_adjclose,
'high': s_high,
'low': s_low,
'volume': s_volume
})
csv_name = 'test_data/result/result-%s_model_%d.csv' % (
result_date.date(), type)
# file_result = open('data/result/' + csv_name, 'rb+')
p_result.to_csv(csv_name)
